With the advent of GenAI, faculty teaching large enrollment courses may find themselves challenged to support students’ true understanding of course concepts. Professor of Public Policy John Hird decided he would continue to develop students’ skills of thinking and writing critically, identifying credible sources, and constructing persuasive arguments, while using chatbots to aid in the process.
What motivated you to include GenAI in your class(es)?
I teach a large introductory Gen Ed course, Controversies in Public Policy, where more than 200 students spend each week tackling a new contentious issue—from nuclear power and immigration to universal basic income and reparations. In past years, students completed traditional two-page policy memos, in which they had to argue a position using evidence and provide a rebuttal to counterarguments. I’m grappling now with trying to improve student learning in a large class in the face of pervasive use of AI chatbots. I’m no fan—AI feels like a sudden, vast, and unplanned experiment to which we didn’t consent. But because today’s GenAI models can produce a polished memo in seconds, and the incentive to cheat is just too high and AI use is undetectable, those memos are no longer workable. I found myself confronting two pressing questions: 1) How do I ensure that students still learn to think and write critically, identify credible sources, construct persuasive arguments, and have their perspectives challenged regardless of their ideology in this new environment? and 2) What kinds of assignments can cultivate those skills in our AI-infused world? As it turned out, AI both created a problem and offered a solution.
What was your approach to bringing GenAI into your course design?
My in-class format has always been intentionally low-tech. Students put away phones, laptops, tablets. Students pick up large notecards at the beginning of class, the only thing on their desks. Several times each session, I pose a question. Students think, write their responses by hand on the notecards, and then we discuss the full range of their views. The exercise also encourages reluctant students to speak since they’ve already formulated their thoughts. I collect and grade the cards and emphasize from day one that the point of meeting together is to interact, engage, and learn from each other. Despite its size, the class is highly participatory.
This year, I kept the phone ban and handwriting requirement, even increasing their weight in the final grade. Rather than trying to prohibit or police the use of generative AI, I redesigned the out-of-class work to incorporate AI deliberately and transparently. Each student now engages in a structured, iterative conversation with an AI that argues the opposite of their initial view on that week’s controversy. It’s a personalized conversation.
For example, if the week’s topic is student-debt relief and a student begins by arguing that all debt should be forgiven for borrowers earning under $100,000, she first writes out her position clearly, summarizing the evidence and logic supporting it. She then embeds her argument into a tailored, detailed prompt that I provide all students, which instructs the AI to take a civil, conversational, but opposing stance—fact-checking the student as needed and raising the strongest counterarguments. The heart of the assignment is what happens next. The student must reply to the AI, either reinforcing her view with additional evidence, refining her position, or revising it altogether. This back-and-forth continues for four rounds. Throughout, each student must justify any claim she makes—just as she would in a traditional memo—but with the added pressure of a responsive, probing interlocutor. After completing the exchange, students write a short (150-word) reflection answering four questions:
- Did your original position change, deepen, or become more nuanced? How?
- Which AI argument or question most challenged your thinking, and why?
- If your view did not change, what in the conversation clarified or strengthened your original stance?
- What additional information (if any) could lead you to alter your position?
They then submit the entire “conversation” for evaluation, which is based on the quality of their arguments, their meaningful engagement with alternative positions, their critical thinking and openness to alternative views, and the thoughtfulness of their reflection.
Students now produce at least as much writing as they did with the old memos, but with richer critical engagement. They learn to craft arguments, deploy evidence thoughtfully, and evaluate the AI’s claims rather than accept them uncritically. I never could have engaged with each student’s views individually in the past, so using the AI to do so at scale is an important advancement. And the process primes them for our Friday TA-led discussions, where they test their refined arguments, and their willingness to listen, in a real, human conversation with classmates. I suspect students are learning more about the material and about themselves, as a result. I remind them that the class structure, the material I present in class, and all grading are prepared and conducted by humans, me and the TAs.
How do you prepare students to use GenAI?
We spend a full week discussing broader issues regarding AI, including its potential positive and negative uses, ethics, environmental impacts, and potential regulations. A couple of students objected to using AI because of its environmental footprint. The build-out of data centers and energy usage is a serious environmental issue, and training the LLMs and heavy use of sophisticated, lengthy video production consumes significant electricity. However, the electricity consumed by simple text-based queries apparently is minimal.
AI chatbots are intentionally designed to be conversational and disarmingly human-like, so I regularly remind students that they are interacting with a machine, not a mind. I also remind them that AI’s persuasively “hallucinate” so they have to fact-check their assertions. Not surprisingly, we encountered a few challenges. Early on, some students used chatbots that ignored instructions, repeated arguments, or failed to respond as genuine interlocutors. Over time, most shifted to Claude, ChatGPT, or Gemini, which consistently provided effective, argument-driven exchanges. And because it’s unusual to require conversations with an AI as part of a course, I explain repeatedly why we are using this approach: it gives each student direct engagement with well-reasoned opposing views and helps them practice critical thinking, argumentation, persuasion, and reflection, skills essential for careers and navigating disagreements beyond the classroom.
What has been the impact of the strategy on student learning?
The most valuable part of this assignment is that students must grapple with smart, evidence-based views that differ from their own, and to think deeply about their own views before engaging in discussion with their peers. Even in a large class with wide-ranging opinions, many students lean left-of-center on many issues and, like most of us, live in ideological bubbles. The AI forces them outside those bubbles. Some of my favorite student comments about the AI conversations are, “What if I can’t come up with any effective counter-arguments?” and “What do I do if the AI convinces me?” Yes, AI systems confidently “hallucinate,” but so do people, which is why students must fact-check the chatbots throughout. The point is not to accept the AI’s authority but to wrestle with it. The students grapple individually with the AI-as-interlocutor, which sets them up for the Friday in-person discussion sections where they test their refined arguments, and occasional newly revised views, with their classmates.
A final advantage is that students can test their ideas without fear of embarrassment. They are mostly 18-20year olds encountering public policy for the first time, and sharing tentative ideas in a large lecture hall or even in smaller sections can feel intimidating. An AI interlocutor offers a low-stakes space to experiment, revise, and think aloud without fear. This private practice helps reluctant students build enough confidence to participate more fully in our real, human discussions with classmates.
What considerations or tips do you have for other faculty interested in having students use GenAI in their courses?
We are all trying to figure out how to promote meaningful learning in an AI-infused world, and what works in my classes may not suit every discipline or setting. Still, the AI-as-interlocutor model has real promise for large courses where instructors want students to confront well-reasoned, individually-tailored alternative views. (For other types of classes, the AI-as-tutor model may work well.) The better AI systems are remarkably effective at this. After the Charlie Kirk murder, several students told me that even though they disagreed with his politics, they sometimes listened to him precisely to encounter opposing perspectives. Many students want to understand alternative viewpoints. With well-designed prompts, AI can provide respectful, evidence-based counterarguments tailored to each student’s starting position, and do so reliably at scale.
Like it or not and for better or worse, AI is here to stay and successive waves of students will have grown up engaging with an AI their entire lives. It’s upending education and it’s up to all instructors to develop ways to promote student learning in this new context.
